Плагины

Панель анализа контекста: посмотрите, из чего состоит контекстное окно модели и как оно меняется — состав относительно размера окна, история по запросам, события сжатия/инъекции и статистика токенов по каждому сообщению.

Profile-wide DSH governance and routing bundle with configurable compaction calls and scoped local semantic memory; compatible with DSH 0.1.0-rc.6 and 0.1.0-rc.7.

DSH session manager: delete with trash/restore/purge, restore archived sessions, recent-activity stats, continue/pause, open log folder, unread markers, fork into a new chat, workspace grouping and sorting, context compaction threshold.

Vision for text-only DeepSeek via Doubao Web by default (zero-cost, no API key — drives your logged-in Chrome through a Windows CDP bridge), with Antigravity IDE quota (flash/pro) or Gemini fallback; auto detail escalation, vision evidence memory with compaction rehydration, content-hash cache, and a bilingual client panel.

Движок сжатия без потерь* и без участия LLM для DeepSeek Harness

История ввода в стиле терминала для веб-composer DeepSeek Harness: стрелки с приоритетом на крайние позиции с точным восстановлением черновика/курсора, история, сохраняемая локально в браузере, обратный поиск Ctrl+R, вызов в рамках рабочего пространства и полностью настраиваемые клавиши — а также учёт скользящего контекста

Импортируйте историю чатов claude-code / codex / reasonix / zcode в сессии dsh: привязка к рабочему пространству, сохранение вызовов инструментов, защита от слишком больших сессий, восстановление сжатия zcode.

Dedicated model routes, tools, and system guidance for vision, compaction, reviews, subagents, titles, and image generation.

Нативный vision-мост провайдера LLM: изображения, вставленные в чат, описываются vision-моделью (Qwen3-VL через pi-ai/llama.cpp), а текстовое описание передаётся текстовой модели DeepSeek для ответа — приём изображений, маршрутизация и сжатие идут через нативные механизмы harness, с LRU-кэшем описаний и повтором при 503.

Vision toolkit for text-only DeepSeek: model-invokable `vision` tool, wrapper adapters for deepseek/opencode-go (v4 flash/pro), Antigravity IDE quota (default, flash/pro) / any OpenAI-compatible VLM / Gemini / local Ollama channels, evidence memory with compaction rehydration, content-hash cache, and a bilingual client panel.

Routes chat images to a fixed OpenAI-compatible vision model, returns factual observations to the selected main model, and reuses session-scoped observations across replay, compaction, and restarts.

Enhanced auto-compaction defaults for DeepSeek Harness agent presets: apply tuned threshold/retention budgets and per-model policies with one command, idempotent with backups and rollback.

Model-authored context pruning for DeepSeek Harness through the official compaction API.

Deterministic context compaction: prunes conversation atoms in reference-graph topological order with no LLM calls during compaction, and keeps pruned atoms recallable by marker.

Deterministic context-pruning compaction backend for dsh (DeepSeek Harness) — zero-LLM summaries, /dcp command, works out of the box. Design references Opencode-DCP/opencode-dynamic-context-pruning.

Read-only performance diagnostics for DeepSeek Harness: session load (open/restore) timing, spill-hit counts, compaction count and trigger, context-injection volume (AGENTS.md/skills/tool-schema token share), and LLM cache hit rate — surfaced via the /fas

Visual percentage and token settings for official DeepSeek Harness compaction

File-based self-evolving memory for DeepSeek Harness v2: compression & extraction over official compaction, daily episodic layer, dream consolidation to semantic files (user/agent/memory/dream + project), catalog disclosure injection, no confidence machin

Pure-incremental directive-driven compaction for DeepSeek Harness: /compact-directive <requirement> summarizes the session middle per your requirement, /trim-directive <requirement> trims the whole conversation per your requirement

The pet that grows with your agent: 5 evolution stages earned from real signals (verified rules, completed sessions, tool calls, compactions), agent-state mirroring, and one-click growth share cards. Zero dependencies, 100% local.

One-command bundle of the three AgentFrame plugins: memory-director, compaction, and aura-scheduler together.

Compaction backend replacing LLM summarization with a deterministic semantic extractor (keeps code/paths/commands, drops chatter) plus 28.4x KV-compression accounting.

Session context triage for long DeepSeek Harness sessions: deterministic screeners trim stale, duplicate, failed, and oversized content through the official ctx.compaction seam.

Focus board for DeepSeek Harness agents: durable, model-maintained notes in the session workspace that pin the objective, constraints, and decisions across compaction and sessions, with automatic context injection, archive on clear, and an optional web panel.